How tools find patterns in flight

In January, Megaputer Intelligence Inc. and Southwest Airlines completed a proof-of-concept demonstration of flight-data analysis and presented their findings to the Global Aviation Information Network, an airline safety community of interest that maintains a Web site at Southwest supplied both structured and unstructured, narrative data from its Aviation Safety Action Program (ASAP).Among the findings was that the tool could visually present patterns of words occurring together in event descriptions submitted by pilots. According to a report on the project, “the results helped Southwest Airlines safety managers locate some potential issues across different aircraft types and airports.”The nonprofit Mitre Corp. of Bedford, Mass., also has built a tool called the Aviation Safety Data Mining Workbench, which it tested last year on data from 12,000 American Airlines pilots’ ASAP reports.According to its report on the test, Mitre gave each ASAP report a unique identifier, then split it into structured data tables—such as aircraft type and category of incident—and unstructured tables consisting of comma-deliminated text from post-event narratives.The software required lots of customization, such as “stemming” similar words to their root for purposes of association. For example, “opens,” “opened,” “opening” and so forth were all stemmed to “open.”Mitre reported that its tool could find patterns such as that 78 percent of Florida takeoffs had an altitude deviation, whereas 55 percent of approaches involved an altitude deviation. In the report, the company stressed that the actual findings on the airline’s data remained confidential.